Reference guide · performance · Published 2026-08-15 · 3 min read
Using PageSpeed Insights
Read a PageSpeed Insights report: field data from CrUX, lab data from Lighthouse, what each score means and its limits.
- ·Field vs lab
- ·Read the report
- ·Limits
What PageSpeed Insights shows
PageSpeed Insights answers one question: is this page fast for real visitors and, if not, why? For any URL it returns two layers. The field layer is real user data from the Chrome User Experience Report (CrUX), reported for the URL's origin and, when enough traffic exists, for the page itself. The lab layer is a fresh Lighthouse run from Google's servers on a single visit. The two answer different questions, which is why the score can disagree.
Field vs lab
| Field (CrUX) | Lab (Lighthouse) | |
|---|---|---|
| Source | Real visits from real Chrome users | One fresh automated run |
| Data | 28-day rolling period, percentiles | A single snapshot |
| CWV values | Actual LCP, INP, CLS at p75 | Estimated from the run |
| Good for | What visitors actually experience | Finding the cause and iterating |
A page can be quick in the lab and slow in the field, or the reverse. The lab run happens from a fast network on a fresh cache, so third-party and server-side network cost do not show the same way. The Core Web Vitals metric page explains the three thresholds you will see here.
How to read the report
Field data
- The three Core Web Vitals, each labelled Good, Needs improvement, or Poor against the thresholds. Good LCP is under 2.5 seconds, Good INP under 200ms, Good CLS under 0.1.
- The value shown is the P75 (75th percentile), meaning three quarters of visits are faster than the number.
- A thumbs-up on "Field data" only means enough mobile traffic exists to report; a thumb-down may be lack of data, not performance.
Lab tables
- Performance score, 0 to 100: a weighted blend of metrics including LCP, INP, CLS, and others. The score is not a real-user measure; it is a lab index for comparing builds.
- Per-metric gauges: each metric has its own measuring-band colour.
- Opportunities: concrete, quantified fixes such as "Eliminate render-blocking resources" and "Reduce main thread work" with estimated savings.
- Diagnostics: pass/fail items like uses efficient image formats, unused CSS, or a long cache TTL.
The limits
- A single lab number masks variability: run it 3 times and expect a spread, then work to the median.
- Field data hides page-level detail until there is traffic. With little data, CrUX reports the origin average.
- Third-party behaviour, network conditions, and server geography all move the lab result. The same URL is a different score in another region.
- The bundle of scored metrics means a site that is fast on LCP but weak on INP can still show a middling score, with the worse metric dragging it.
Using it as a workflow
Run PageSpeed Insights per URL at a regular cadence (before and after a release) and capture the lab score plus the top three opportunities. The 30 minute performance audit tutorial turns this report into a list of fixes, and the LCP route there shows the elements a real improvement creates.